Rivers across the Siberian Arctic unearth the patternsof carbon release from thawing permafrostBirgit Wilda,b,1, August Anderssona, Lisa Brödera,b,c, Jorien Vonkc, Gustaf Hugeliusb,d, James W. McClellande,Wenjun Songf, Peter A. Raymondf, and Örjan Gustafssona,b,1

aDepartment of Environmental Science and Analytical Chemistry, Stockholm University, 106 91 Stockholm, Sweden; bBolin Centre for Climate Research, StockholmUniversity, 106 91 Stockholm, Sweden; cDepartment of Earth Sciences, Vrije Universiteit Amsterdam, 1081 HV Amsterdam, The Netherlands; dDepartment ofPhysical Geography, Stockholm University, 106 91 Stockholm, Sweden; eMarine Science Institute, University of Texas at Austin, Port Aransas, TX 78373; and fYaleSchool of Forestry and Environmental Studies, New Haven, CT 06511

Edited by Mark H. Thiemens, University of California, San Diego, La Jolla, CA, and approved March 26, 2019 (received for review July 9, 2018)

Climate warming is expected to mobilize northern permafrost andpeat organic carbon (PP-C), yet magnitudes and system specifics ofeven current releases are poorly constrained. While part of thePP-C will degrade at point of thaw to CO2 and CH4 to directly am-plify global warming, another part will enter the fluvial network,potentially providing a window to observe large-scale PP-C remobi-lization patterns. Here, we employ a decade-long, high-temporalresolution record of 14C in dissolved and particulate organic carbon(DOC and POC, respectively) to deconvolute PP-C release in the largedrainage basins of rivers across Siberia: Ob, Yenisey, Lena, andKolyma. The 14C-constrained estimate of export specifically fromPP-C corresponds to only 17 ± 8% of total fluvial organic carbonand serves as a benchmark for monitoring changes to fluvial PP-Cremobilization in a warming Arctic. Whereas DOC was dominatedby recent organic carbon and poorly traced PP-C (12 ± 8%), POCcarried a much stronger signature of PP-C (63± 10%) and representsthe best window to detect spatial and temporal dynamics of PP-Crelease. Distinct seasonal patterns suggest that while DOC primarilystems from gradual leaching of surface soils, POC reflects abruptcollapse of deeper deposits. Higher dissolved PP-C export by Oband Yenisey aligns with discontinuous permafrost that facilitatesleaching, whereas higher particulate PP-C export by Lena andKolyma likely echoes the thermokarst-induced collapse of Pleisto-cene deposits. Quantitative 14C-based fingerprinting of fluvial or-ganic carbon thus provides an opportunity to elucidate large-scaledynamics of PP-C remobilization in response to Arctic warming.

carbon cycle | climate change | radiocarbon | peat | leaching

The destabilization of permafrost and peat deposits in awarming Arctic involves a range of mechanisms that act on

different temporal and spatial scales. Rising temperatures pro-mote a gradual deepening of the seasonally thawed active layer atthe surface of permafrost soils and a decrease in areal permafrostextent at the southern margin of the permafrost zone (1). Risingtemperatures and increasing precipitation can further induceabrupt landscape collapse and degradation of deeper organiccarbon deposits. Ice-rich permafrost deposits are particularly vul-nerable to collapse (thermokarst) (2), including Holocene but alsoPleistocene deposits that are still widespread, especially acrossnortheastern Siberia (Ice Complex deposits, or Yedoma) (3).Changing climatic conditions might further destabilize deep peatdeposits that have accumulated during the Holocene across thecircum-Arctic (4–6). Peat is particularly abundant in the World’slargest wetland—the West Siberian Lowland—which is largelyunderlain by vulnerable discontinuous permafrost and projected toexperience a further decrease in permafrost extent (7).Permafrost and peat degradation affects vast and remote areas

where there is very limited access to field data. To complementexisting, rare, and largely point-specific studies across the het-erogeneous landscape of the Siberian Arctic and tackle theupscaling challenge, this study employs rivers as natural integra-tors of carbon mobilization because they transport organic carbonreleased by abrupt collapse and erosion of old Holocene and

Pleistocene deposits, as well as organic carbon leached from deep-ening active layers, next to organic carbon recently fixed by plants.The different 14C ages of these organic carbon sources are used in afingerprinting approach to distinguish flux components from differentpermafrost and peat organic carbon (PP-C) pools versus recentprimary production.We take advantage of the unique decade-long, high temporal

resolution records of organic carbon fluxes and 14C contents inthe Ob, Yenisey, Lena, and Kolyma [2003–2013; n = 110 for par-ticulate organic carbon (POC), n = 137 for dissolved organic carbon(DOC) covering all seasons] generated by the river monitoringprograms Pan Arctic River Transport of Nutrients, Organic Matter,and Suspended Sediments (PARTNERS) and Arctic Great RiversObservatory (ARCTIC-GRO) (8). The basins of the four riversspan 110° in longitude and cover a combined area of 8.2 million km2,with 5.8 million km2 in the northern permafrost region. This areacorresponds to 26% of the northern circumpolar permafrost areaand 40% of the northern Eurasian permafrost area (Fig. 1). Com-bining the 14C datasets of both POC and DOC with an extensivedatabase on 14C fingerprints of the potential organic carbon sourcesusing statistical source apportionment, this study provides aquantitative estimate of fluvial organic carbon export specifically

Significance

High-latitude permafrost and peat deposits contain a largereservoir of dormant carbon that, upon warming, may partlydegrade to CO2 and CH4 at site and may partly enter rivers.Given the scale and heterogeneity of the Siberian Arctic,continent-wide patterns of thaw and remobilization have beenchallenging to constrain. This study combines a decade-longobservational record of 14C in organic carbon of four largeSiberian rivers with an extensive 14C source fingerprint data-base into a statistical model to provide a quantitative parti-tioning of the fraction of fluvially mobilized organic carbonthat specifically stems from permafrost and peat deposits, andseparately for dissolved and particulate vectors, across theSiberian Arctic, revealing distinct spatial and seasonal systempatterns in carbon remobilization.

Author contributions: B.W., L.B., J.V., G.H., J.W.M., P.A.R., and Ö.G. designed research;B.W., J.W.M., W.S., and P.A.R. performed research; B.W. and A.A. analyzed data; andB.W., A.A., L.B., J.V., G.H., J.W.M., P.A.R., and Ö.G. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY).

Data deposition: All data used in this study are available in the Supplementary Informa-tion and have been deposited in Stockholm University’s Bolin Centre Database (https://bolin.su.se/data/wild-2019). The MATLAB code is available at https://git.bolin.su.se/bolin/wild-2019.1To whom correspondence may be addressed. Email: birgit.wild@aces.su.se or orjan.gustafsson@aces.su.se.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1811797116/-/DCSupplemental.

Published online May 6, 2019.

10280–10285 | PNAS | May 21, 2019 | vol. 116 | no. 21 www.pnas.org/cgi/doi/10.1073/pnas.1811797116

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 22

, 202

0

source-apportioned to permafrost and peat deposits and assessesthe potential of 14C signatures of POC and DOC to monitorchanges in PP-C release in a warming climate.

Results and DiscussionPreaged Organic Carbon in Siberian Rivers. All four rivers showsignificantly lower Δ14C values for POC than for DOC (flux-weighted t test, P < 0.001; Fig. 2), signaling a greater pro-portion of old carbon in POC, presumably from permafrost andpeat deposits (9–11). This observation is in line with previousstudies that suggest erosion of deep and consequently old de-posits along riverbanks as a large source of POC, but not DOC,to Arctic rivers (12). Across rivers and seasons, POC-Δ14C valuesaveraged −261 ± 82‰ (flux-weighted mean ± SD, n = 110), cor-responding to mean conventional 14C ages of 2,400 (SD +900/−800) y.The POC-Δ14C values of the eastern-most river, Kolyma, were sig-nificantly lower than those of the Lena and of the western rivers,Ob and Yenisey (Fig. 2 and SI Appendix, Table S7). This pattern isconsistent with a systematic increase in 14C ages from west to eastalong the Eurasian–Arctic continental border that has been pre-viously described for river POC and river delta surface sediments (10,13, 14) and suggests a greater influence of compartments with highreservoir ages such as the Ice Complex deposits toward the east.In contrast, more contemporary Δ14C values of DOC indicate

a relatively higher contribution of recently fixed organic carbon.Across all rivers and seasons, DOC-Δ14C values averaged +72 ±39‰ (flux-weighted mean ± SD, n = 137), pointing at a domi-nance of carbon taken up by plants between the 1950s and thepresent, when atmospheric CO2 was enriched in 14C due to nu-clear weapons tests (15). In contrast to POC, DOC-Δ14C valueswere lowest in the Ob, followed by the Kolyma and Yenisey aswell as the Lena (Fig. 2 and SI Appendix, Table S7). The DOC-Δ14C values further decreased significantly from spring (icebreakup; May and June) to summer/fall (ice-free period; Julythrough October) and winter (ice-covered period; Novemberthrough April) in all four rivers.

Quantifying the Permafrost and Peat Component of Fluvial CarbonExport. While the 14C signatures of POC and DOC in Arcticrivers can give some indication of carbon release from high-latitudepermafrost and peat deposits, pinpointing and monitoring thespatial and temporal dynamics of PP-C release is challenging be-cause both POC and DOC represent mixtures of carbon from dif-ferent sources. Here, isotope-based source apportionment withMarkov chain Monte Carlo simulations provides the tool to quan-tify the relative contribution of PP-C to the total fluvial organic

carbon load and thereby isolate the spatial and temporal patterns ofPP-C release. This approach allows us to calculate the contributionof different organic carbon sources to river DOC and POC usingtheir Δ14C values while accounting for variability in both river ob-servations and source endmember values (16). The Δ14C values ofrecent organic carbon and three potential sources of preaged PP-Cwere estimated based on an extensive literature review that is de-scribed in detail in SI Appendix, Supplementary Information Text.The Δ14C endmember of recent terrestrial primary production wasconstrained based on observations from litter and organic layers to+97 ± 125‰ (n = 58; Fig. 2), reflecting the contemporary to onlydecades-aged nature of this carbon pool. The three potential pre-aged PP-C sources were (i) the active layer with average Δ14Cvalues of −198 ± 148‰ (1,700 y, n = 60); (ii) Holocene permafrost,peat, and thermokarst deposits with Δ14C values of −568 ± 157‰(6,700 y, n = 138); and (iii) Pleistocene permafrost deposits such asIce Complex deposits with Δ14C values of −955 ± 66‰ (24,800 y,n = 329). For the latter two carbon sources, we specifically con-sidered only exposures along riverbanks and coasts to most re-alistically represent the Δ14C range of material that may enter riversby erosion. Although the high turbidity in the rivers constrainsphotosynthesis to a thin layer of surface waters (17), aquatic pro-duction by phytoplankton and bacteria can potentially contribute tothe total POC load (18). Nevertheless, the mineralization of ter-restrial carbon during river transport typically leads to oversaturationin CO2 compared with the atmosphere and, thus, to limited influx ofatmospheric CO2 (19–23). Aquatic production is consequentlylargely fueled by recycling of terrestrially derived carbon andtherefore not considered an independent carbon source here(see also discussion in Fate of PP-C During River Transport).Source apportionment between the recently formed carbon

reservoirs versus the preaged PP-C was performed in three sce-narios assuming different contributions of active layer, Holocenedeposits, and Pleistocene deposits to the PP-C flux (see SI Ap-pendix, Supplementary Information Text for details). The BestEstimate scenario represents the most realistic and conservativeestimate because it assumes a contribution of all PP-C com-partments to the PP-C flux and considers the uncertainties of notonly individual carbon source Δ14C values but also their pro-portional contribution to the PP-C flux. Pleistocene deposits wereconsidered only for the Lena and Kolyma catchments where theseare abundant (Fig. 1). The sensitivity of the model’s results to theassumptions of the Best Estimate scenario was additionally testedin the Maximum and Minimum scenarios assuming a contributionof only the youngest (active layer) or oldest (Ob and YeniseyHolocene deposits, Lena and Kolyma Pleistocene deposits) PP-Ccompartment, respectively. These scenarios provide an even moreconservative uncertainty envelope of the estimated PP-C flux, butare realistic only on very small spatial scales and not on the largescales of the great Siberian rivers that integrate carbon releasefrom various sources within their catchments.The fluvial PP-C export vector was calculated by combining

the 14C-constrained contribution of PP-C to total fluvial organiccarbon (SI Appendix, Table S8) with previous estimates of the totalfluvial organic carbon flux load (10, 24) (SI Appendix, Table S9).In the Best Estimate scenario, we thus arrive at a (DOC + POC)combined PP-C export of 3.0 ± 0.3 Tg PP-C per year by Ob,Yenisey, Lena, and Kolyma (Minimum: 2.0 ± 0.2 Tg y−1; Maxi-mum: 5.0 ± 0.5 Tg y−1; Table 1). This first estimate of thedeconvoluted PP-C export component for the four largest Siberianrivers represents a benchmark for monitoring the fluvial remobi-lization of preaged organic carbon from permafrost and peat de-posits in a warming climate, and contributes toward a three-dimensional, quantitative understanding of Arctic carbon cyclingthat considers not only vertical but also lateral fluxes. Our estimatedemonstrates that the PP-C component corresponds to only 17 ±8% of the total fluvial organic carbon load of 17.0 ± 1.3 Tg y−1(Minimum: 12 ± 6%; Maximum: 29 ± 11%; Fig. 3A). Hence, thisstudy reveals that patterns of fluvial PP-C remobilization are maskedby large fluxes of recent organic carbon; quantitatively dissecting thePP-C flux, however, opens an observational window to monitor PP-C

Fig. 1. Northeastern Eurasia with watershed margins of the Ob, Yenisey,Lena, and Kolyma rivers, underlain with the spatial extents of continuous,discontinuous, sporadic, and isolated permafrost (60), as well as of Pleisto-cene Ice Complex deposits (61).

Wild et al. PNAS | May 21, 2019 | vol. 116 | no. 21 | 10281

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 22

, 202

0

release across heterogeneous landscapes and seasons and therebyadvance our understanding of PP-C vulnerability.Although fluvial organic carbon was strongly dominated by

DOC (90 ± 4%), all three model scenarios suggest that POCcontributed about one-third to one-half of the total PP-C export(Best Estimate: 38 ± 19% of PP-C export in the form of POC;Fig. 3B and SI Appendix, Fig. S2). While DOC contained mostlyrecent organic carbon and only 12 ± 8% PP-C, POC carried astrong PP-C 14C signature (63 ± 10% of POC from PP-C; Fig.3A). Hence, the 14C signature of POC offers the best opportunityto chronicle the dynamics of PP-C remobilization.

Unearthing the Seasonal and Spatial Dynamics of PP-C Remobilization.Isotopically deconvoluting recent organic carbon and PP-C con-tributions to fluvial carbon fluxes provides the opportunity to studythe seasonal and spatial dynamics specifically of PP-C in riverswithout interference by the much larger flux of recent organiccarbon. The PP-C in dissolved and particulate carrier phasesshowed distinct seasonal patterns that suggest different dominantmobilization pathways (Fig. 3C). The contribution of PP-C toDOC increased from spring to summer/fall and winter (SI Ap-pendix, Fig. S3 and Table S8), which points to gradual leaching asa key mechanism of dissolved PP-C release. Gradual leachingincreasingly mobilizes deeper (and older) organic carbon aftertop-down soil thaw from spring to summer and refreezing fromfall to winter (25, 26), including carbon thawed at the permafrosttable by active-layer deepening. Seasonal differences in dissolvedPP-C release might also be related to changes in the relative im-portance of surface runoff and deep groundwater-dominated flowpaths in the river catchments (27). Overall, the remobilization ofPP-C in dissolved form was delayed compared with that of recentcarbon. In contrast, the contribution of PP-C to total POCremained constantly high throughout the year, and no distinctdifferences in seasonal dynamics of PP-C versus recent carbonwere observed (SI Appendix, Table S8). Particulate PP-C mighttherefore, to a larger extent, derive from abrupt collapse of deepHolocene and/or Pleistocene deposits that can instantaneouslyrelease organic carbon of various ages into aquatic systems(see refs. 28 and 29); for example, by thermal or ice jam-inducedbank erosion (30). Contrasting DOC and POC mobilizationpathways have been previously suggested based on the different14C ages of DOC and POC (11, 31); isolating the PP-C componentof DOC and POC and pinpointing its seasonal patterns finallyconfirms gradual leaching and abrupt collapse as the mainsources of dissolved and particulate PP-C release, respectively.Overall, the export of recent organic carbon was dominated by thespring freshet, with 52 ± 13% of the annual export in spring, 39 ±13% in summer/fall, and 10 ± 3% in winter. These findings are inline with previous observations of high lignin concentrations in

DOC in spring that suggest strong leaching of recent plant litterduring this period (9). The export of PP-C, in contrast, was shiftedtoward later seasons, with 38 ± 12% of the annual flux in spring,46 ± 12% in summer/fall, and 16 ± 9% in winter.Differences in the export of dissolved and particulate PP-C

also between rivers provide an opportunity to assess the vulnera-bility of high-latitude carbon stocks across geospatial climaticgradients. Export of PP-C in Ob and Yenisey was dominated byDOC (Fig. 3B). Leaching of organic carbon from soils to streamsmay be facilitated here by discontinuous permafrost in the Ob andYenisey catchments (Fig. 1) (14), as well as by large peatlands thatprovide limited retention of organic carbon by sorption to soilminerals (4, 32). In contrast, a larger fraction of PP-C export byLena and Kolyma was in particulate form (Fig. 3B), reflectingstronger erosion, likely to a large extent of Pleistocene Ice Com-plex deposits that are abundant in both catchments and particu-larly susceptible to erosion due to their high ice content (2, 3).Overall, the contribution of PP-C to the total fluvial carbon loadwas lowest in the Yenisey and increased both toward the Lena andKolyma in the east and toward the Ob in the west. The Ob showedthe strongest PP-C contribution to total export, had the highestrate of PP-C flux, and accounted for 40% of the PP-C exported bythe four large Siberian rivers (Table 1).

Fate of PP-C During River Transport. Comparing Δ14C and δ13Cvalues of fluvial organic carbon and its deduced sources mayfurther provide information about the fate of PP-C during rivertransport. The δ13C values of terrestrial carbon pools in thedrainage area of the four rivers fall within a narrow range char-acteristic of dominant C3 vegetation, with estimates of −27.7 ±1.3‰ for terrestrial primary production, −26.4 ± 0.8‰ for the

Table 1. Estimates of total organic carbon (DOC + POC) fluxes inOb, Yenisey, Lena, and Kolyma, as well as carbon fluxes frompermafrost and peat deposits

RiverTotal organic carbon*

(Tg y−1)

PP-C† (Tg y−1)

Best Estimate Minimum Maximum

Ob 4.7 ± 0.7 1.2 ± 0.2 0.9 ± 0.1 1.8 ± 0.3Yenisey 4.9 ± 0.4 0.7 ± 0.2 0.5 ± 0.1 1.1 ± 0.2Lena 6.5 ± 1.0 0.9 ± 0.2 0.5 ± 0.1 1.7 ± 0.3Kolyma 0.9 ± 0.2 0.2 ± 0.0 0.1 ± 0.0 0.3 ± 0.1All rivers 17.0 ± 1.3 3.0 ± 0.3 2.0 ± 0.2 5.0 ± 0.5

Values are represented as mean ± SD.*Estimates from refs. 10 and 20.†Best Estimate, Minimum, and Maximum represent model scenarios for PP-Cfluxes. See SI Appendix, Table S9 for detailed data.

Fig. 2. Carbon isotopic composition of DOC (Left) andPOC (Center) in four large Siberian rivers, expressed asΔ14C and δ13C values. Boxplots show medians with 25thand 75th percentiles as box limits and 10th and 90thpercentiles as whiskers, with isotopic data weighted bythe flux rate of DOC and POC at the corresponding timepoint. Endmember Δ14C and δ13C values (mean ± SD) ofpotential organic carbon sources (Right) indicate recentterrestrial primary production (Terr. prod.), active layer,Holocene deposits (Hol. dep.), and Pleistocene deposits(Pleist. dep.); note that δ13C values for Holocene de-posits are not shown here; see SI Appendix, Supple-mentary Information Text for details. Conventional 14Cdates derived from Δ14C values are indicated at theright margin. BP, before present. Both Δ14C and δ13Cvalues were significantly lower for POC than for DOC(flux-weighted t test, P < 0.001; see SI Appendix, TableS7 for statistical analyses of differences between riversand seasons).

10282 | www.pnas.org/cgi/doi/10.1073/pnas.1811797116 Wild et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 22

, 202

0

active layer, and −26.3 ± 0.7‰ for Pleistocene Ice Complex de-posits based on previous studies (see SI Appendix, SupplementaryInformation Text for details). The δ13C values of river DOC werein the same range (−27.0 ± 2.0‰, flux-weighted mean ± SD of allrivers and seasons, n = 134; Fig. 2), whereas the δ13C values ofriver POC were significantly lower, with an annual mean of−28.7 ± 1.6‰ (n = 162) and a wintertime mean of −31.1 ± 3.2‰(n = 37) despite a predominantly terrestrial 14C signature.Differences in the 13C content of POC compared with its

source indicate loss of POC during transport. The 13C content oforganic carbon is therefore not a conservative source markerduring long-range aquatic transport. Fluvial organic carbon maybe mineralized to dissolved inorganic carbon, which is partlyoutgassed to the atmosphere and partly taken up by phyto-plankton to fuel photosynthesis. Terrestrial carbon may therebyreenter the POC pool in the form of phytoplankton after anadditional 13C fractionation step, consequently lowering POC δ13C values. In line with the recycling of organic carbon during rivertransport, Ob and Yenisey phytoplankton shows low δ13C values(−30.6 ± 3.3‰, n = 24; see SI Appendix, Supplementary In-formation Text and ref. 33) similar to those of POC. Lower δ13Cvalues of phytoplankton have also been observed in the LaptevSea (which receives strong carbon input from the Lena River) ascompared to the East Siberian Sea (34). Degradation of terres-trially derived organic carbon during river transport is furthersupported by previous incubation studies that show a rapid lossof DOC in permafrost leachates and rivers (35–43), with highestlosses of up to 53% within 9 d for DOC from Ice Complex de-posits (36, 37, 44). Comparable incubation studies on thedegradability of POC are urgently needed, considering the highcontribution of POC to PP-C export as demonstrated in thisstudy. In addition to the “fast pool” within terrestrially derivedcarbon, other fractions will be degraded more slowly [as in-dicated by CO2 oversaturation of river and coastal ocean watersfar away from the original source location compared with theatmosphere (23, 45)], or will be resequestered by sedimentation.Applying 14C-based source apportionment to DOC and POC atdifferent locations along the fluvial network in future studiescould improve our understanding of terrestrial carbon transfer torivers, the fraction of DOC and POC degraded or sedimentedduring different stages of aquatic transport, and thereby the“boundless carbon cycle” that connects multiple cooccurringprocesses on land, in the water, and in the atmosphere.

The 14C Signature of Fluvial Organic Carbon as Indicator of Future PP-C Release.The 14C signature of carbon in large rivers may serve asan indicator to monitor the release of carbon from high-latitude

permafrost and peat deposits in a warming climate. The sensi-tivity of fluvial Δ14C values to changes in PP-C release was testedby simulating a decrease or increase in PP-C flux by factorsranging from 0.5 to 2.0. The Best Estimate approach, at constantrecent carbon flux, was used for this sensitivity test. The resultingshift in Δ14C values of fluvial organic carbon depended onbaseline Δ14C values (i.e., Δ14C values without change in PP-Cflux), with highest sensitivity at baseline Δ14C values between−100‰ and −200‰ in the western rivers Ob and Yenisey, andbetween −100‰ and −300‰ in the eastern rivers Lena andKolyma (SI Appendix, Fig. S4). Doubling the PP-C flux resultedin a maximum decrease in Δ14C by 82‰ (western rivers) and109‰ (eastern rivers). Halving the PP-C flux resulted in anincrease by the same values.The suitability of DOC and POC to monitor changes in PP-C

release in a warming Arctic depends on (i) current DOC-Δ14Cand POC-Δ14C values, including their variability; (ii) the sensi-tivity of Δ14C baseline values to changes in PP-C release; and(iii) the expected change in PP-C release in dissolved and par-ticulate form. The sensitivity of the DOC pool to resolve changesin PP-C release is challenged by the strong dilution of PP-C withrecent carbon in this pool. For Yenisey and Lena, increases inPP-C release by an additional 159% and 130%, respectively,would be required to result in a statistically significant difference(P < 0.05) to baseline DOC-Δ14C values (SI Appendix, Fig. S4).Sensitivity is higher in Ob and Kolyma, with minimum resolvableincreases in PP-C release by 27% and 48%, respectively. Con-sidering that gradual active-layer leaching releases mostly dis-solved PP-C as indicated by the spatial and temporal dynamics ofthe isotopically deconvoluted PP-C flux, our findings suggestoverall low sensitivity of the fluvial 14C signature to active-layerdeepening. However, the sensitivity of DOC-Δ14C might be suf-ficient in Ob and Kolyma, as well as in winter when the relativecontribution of PP-C reaches its annual maximum and when, es-pecially, the deep active layer may still be unfrozen (25, 26).By contrast, the 14C signature of POC is likely a sensitive in-

dicator of abrupt collapse of deeper PP-C deposits. ParticulatePP-C stemmed, to a large extent, from erosion and stronglydominated the total POC flux (63 ± 10% PP-C), resulting in ahigher sensitivity of POC-Δ14C values compared to DOC-Δ14Cvalues (SI Appendix, Fig. S4). An increase in PP-C release by only24 to 36% would thus induce statistically resolvable changes inPOC-Δ14C in all four rivers (SI Appendix, Fig. S4). In compari-son, the total fluvial sediment load by erosion has been projectedto increase by 30 to 122% until 2100 in the six largest Eurasianrivers, including Ob, Yenisey, Lena, and Kolyma (46), and by200 to 600% in a smaller river in the Canadian Arctic (47).Both projections are likely minimum estimates because they

A B C

Fig. 3. Comparison of recent organic carbon and PP-C fluxes. (A) Contribution of PP-C to total organic carbon export in four large Siberian rivers. Fractions ofPP-C compared with recent primary production are based on the Best Estimate scenario; the shaded areas indicate the intervals between the Minimum andMaximum scenarios. (B) Contribution of carbon from recent primary production and PP-C in dissolved and particulate form to the total organic carbon exportin individual rivers (Best Estimate scenario; see SI Appendix, Table S9 for other scenarios). (C) Contribution of DOC and POC in spring, summer/fall, and winterto the annual export of total organic carbon, recent organic carbon, and PP-C (Best Estimate scenario; see SI Appendix, Fig. S2 for other scenarios). Totalorganic carbon export was quantified based on discharge, POC, and DOC concentration measurements using the LOADEST program (10, 24).

Wild et al. PNAS | May 21, 2019 | vol. 116 | no. 21 | 10283

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 22

, 202

0

considered only the changes in general hydrological propertiesand temperature, not the expected increases in permafrost thawand thermokarst formation in the river catchments (2). Changesmight be even larger if tipping points in Arctic permafrost thawingare passed, as indicated in a recent study (48). Our analyses sug-gest the high sensitivity of POC-Δ14C values to even comparativelymoderate increases in PP-C release by erosion in the rivercatchments; the Δ14C values of POC consequently provide thebest observational window to detect changes in PP-C release.

ConclusionsRising temperatures are expected not only to accelerate thedegradation of PP-C to CO2 or CH4 at point of thaw, therebyinducing a positive feedback to climate warming (49), but also,simultaneously, to alter carbon release into the fluvial network(32, 50). Part of the fluvial carbon will eventually reach theArctic Ocean where a significant portion is degraded, thousandsof kilometers away from point of thaw (51), giving rise to CO2efflux to the atmosphere and severe ocean acidification (45).However, large spatial heterogeneity in organic carbon thaw andremobilization is expected on local, drainage basin, and sub-continental scales (1, 2, 32), implying that complementary ap-proaches to site-specific observations are needed to meet thisupscaling challenge.This study combines an extensive, decade-long set of river ob-

servations that integrates release processes over large drainagebasins with 14C-based source apportionment to provide quantita-tive constraints on organic carbon mobilization specifically frompermafrost and peat deposits across Siberia. Despite the vast extentof old permafrost and peat deposits in the catchments of Ob,Yenisey, Lena, and Kolyma, fluvial organic carbon and, in partic-ular, DOC were strongly dominated by recent primary production.Hence, DOC in Arctic rivers carries limited information on per-mafrost carbon release. In contrast, POC was dominated byremobilized PP-C (63 ± 10%). Although POC constituted merely10% of the total fluvial organic carbon load, it thus accounted formore than a third of the fluvial PP-C export and represents the bestobservational window to monitor PP-C release in a warming cli-mate. Deconvoluting the relative contributions of recent organiccarbon versus PP-C fluxes revealed distinct seasonal patterns, withcarbon export from recent vegetation dominated by the springfreshet and with PP-C mobilization shifted toward summer, fall,and winter, highlighting the importance of late-season PP-C releaseprocesses. Dissolved PP-C export dominated the western rivers Oband Yenisey whose drainage basins are characterized by less per-mafrost coverage and less soil carbon retention. In contrast, higherparticulate PP-C export in the eastern rivers Lena and Kolymaechoes the thermokarst-induced, abrupt collapse of Pleistocene IceComplex deposits. Quantitative 14C-based fingerprinting of fluvialorganic carbon, especially of the POC, thus provides informationon changes to the otherwise invisible subsurface cryosphere carbonover large spatial scales in response to Arctic warming and ad-vances our understanding of the inner workings of large-scalepermafrost carbon remobilization—an essential component formeaningful predictions of the Arctic PP-C–climate feedbacks.

MethodsDOC and POC Sampling and Analyses. Samples for DOC and POC analyses werecollected at Salekhard (Ob), Dudinka (Yenisey), Zhigansk (Lena), and Cherskiy(Kolyma) between July 2003 and November 2013 as part of the PARTNERSand ARCTIC-GRO programs (8). Details on DOC and POC sampling andanalysis methods can be found in SI Appendix, Supplementary InformationText, as well as in previously published papers (10, 24, 52) and in the meta-data provided with the publicly available datasets (https://arcticgrea-trivers.org/). Original data are presented in SI Appendix, Table S6 and on theproject homepage (https://arcticgreatrivers.org/), and have been used forother purposes in previous studies (9, 10, 52).

Data were categorized into a spring period (May and June), characterizedby ice breakup and high discharge; a combined summer and fall period (Julyto October), during which rivers are ice-free; and a winter period (Novemberto April), when rivers are ice covered and discharge is low (10, 24). In thecase of DOC-Δ14C, the spring season included 59 individual data points,

the summer/fall included 49, and the winter included 29. In the case ofPOC-Δ14C, spring was represented by 50 data points, summer/fall by 42,and winter by 20. We thereby also captured the spring freshet, whendischarge and carbon export can change rapidly at high temporal res-olution (SI Appendix, Fig. S1). Total DOC and POC export rates in indi-vidual seasons and rivers were derived from previous publications basedon the PARTNERS/ARCTIC-GRO data on discharge, POC, and DOC con-centrations. These studies applied the US Geological Survey Load Estimator(LOADEST) program that builds regression equations to relate variations in fluxwith variations in discharge over multiple years while accounting for seasonalvariations in those relationships. A detailed description of the approach isprovided in the original publications (10, 24). Isotopic data were weighted bythe corresponding DOC or POC fluxes at the time of sampling to achieve aproportional representation of periods with high and low DOC and POC flux.Percentiles of flux-weighted data for box plots were calculated using the Hmiscpackage (53) in R 3.5.1 (54). Differences between DOC and POC were tested forsignificance using t tests of flux-weighted data in the weights package (55) in R,and differences between rivers and seasons of flux-weighted data were testedusing two-way ANOVA with Tukey’s honest significant difference post hoc testin the HH package (56) in R.

Source Apportionment and Markov Chain Monte Carlo Simulations. The con-tribution of PP-C to DOC and POC was quantified based on the Δ14C sig-natures of four potential organic carbon source pools that were derivedfrom extensive literature review as described in detail in SI Appendix, Sup-plementary Information Text, substantially updating previous versions ofendmember databases from the Siberian Arctic (57, 58): (i) Recent terrestrialprimary production Δ14C values were estimated as 97.0 ± 124.8‰, based onpublished data from organic and litter layers in arctic, subarctic, and borealsystems in northern Russia, northern Scandinavia, northern Canada, andAlaska (n = 58); (ii) Active-layer Δ14C values were constrained as −197.5 ±148.4‰, based on data from active layers and nonpermafrost soils (excludingorganic layers) in northern Siberia (n = 60); (iii) Holocene permafrost, peat,and thermokarst deposits were of similar age and thus combined; mean Δ14Cvalues of Holocene deposits were calculated from exposures of peat andthermokarst in northern Siberia as −567.5 ± 156.7‰ (n = 138); and (iv)Pleistocene deposit Δ14C values were estimated as −954.8 ± 65.8‰, based ondata from Pleistocene Ice Complex deposit exposures in northeasternSiberia (n = 329). Considering the common oversaturation of Arctic riverswith CO2 and the consequently limited influx of atmospheric CO2 (19–23),aquatic primary production represents largely recycling of terrestrial car-bon and is thus not considered an independent organic carbon source.Nevertheless, the Δ14C range of atmospheric CO2 during the time ofsampling of +48 ± 11‰ (59) falls within the range of recent terrestrialprimary production; any potential minor influx of atmospheric CO2

would therefore be within the uncertainty of the endmember forrecent carbon.

Source apportionment was performed for two endmembers that repre-sent recent carbon and PP-C (see SI Appendix, Supplementary InformationText for details). The Δ14C values of the PP-C endmember were calculated forthree scenarios assuming different contributions of organic carbon fromactive layers, Holocene deposits, and Pleistocene deposits to the PP-C flux.The Best Estimate scenario likely represents the most realistic estimate be-cause it considers a contribution of all compartments to the PP-C flux. Spe-cifically, a least-biased approach was used in which all fractional combinationsof individual compartments are set to be equally likely. Compared with theassumption of equal contributions, this approach results in the same averageΔ14C value of the combined PP-C endmember, but also in a larger uncertaintythat includes both the uncertainty of individual endmember constraints andthe uncertainty of their proportional contribution to the PP-C endmember.Pleistocene deposits were considered only for the Lena and Kolyma catch-ments where they are abundant (Fig. 1). Model sensitivity was tested in theMaximum and Minimum scenarios assuming a contribution of only theyoungest (active layer) or oldest (Ob and Yenisey Holocene deposits, Lena andKolyma Pleistocene deposits) PP-C compartment, respectively. To account forthe variability of the endmembers, the relative source contribution estimateswere calculated within a Bayesian Markov chain Monte Carlo framework (16).The simulations were run in MATLAB (version 2014b), using 1,000,000 itera-tions, a burn-in phase of 10,000, and a data thinning of 10.

Data Availability. All data used in this study are available in the Supple-mentary Information and have been deposited in Stockholm University’sBolin Centre Database (62). The MATLAB code used for statistical sourceapportionment is available at https://git.bolin.su.se/bolin/wild-2019 (63).

10284 | www.pnas.org/cgi/doi/10.1073/pnas.1811797116 Wild et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 22

, 202

0

ACKNOWLEDGMENTS. This study was funded by the US National ScienceFoundation (Grants 0229302, 0732821, and 1107774), the Swedish Re-search Council Vetenskapsrådet [Contracts 621-2013-5297 and 2017-01601(to Ö.G.) and 330-2014-6417 (to G.H.)], and the European Research Council

[ERC-AdG CC-TOP Project 695331 (to Ö.G.) and ERC-StG THAWSOME Proj-ect 676982 (to J.V.)]. Ö.G., B.W., J.V., and G.H. additionally acknowl-edge the European Union Horizon 2020 project Nunataryuk (Contract773421).

1. Koven CD, et al. (2011) Permafrost carbon-climate feedbacks accelerate globalwarming. Proc Natl Acad Sci USA 108:14769–14774.

2. Olefeldt D, et al. (2016) Circumpolar distribution and carbon storage of thermokarstlandscapes. Nat Commun 7:13043.

3. Strauss J, et al. (2013) The deep permafrost carbon pool of the Yedoma region inSiberia and Alaska. Geophys Res Lett 40:6165–6170.

4. Frey KE, Smith LC (2005) Amplified carbon release from vast West Siberian peatlandsby 2100. Geophys Res Lett 32:L09401.

5. Alexandrov GA, Brovkin VA, Kleinen T (2016) The influence of climate on peatlandextent in Western Siberia since the Last Glacial Maximum. Sci Rep 6:24784.

6. Macdonald GM, et al. (2006) Rapid early development of circumarctic peatlands andatmospheric CH4 and CO2 variations. Science 314:285–288.

7. Chadburn SE, et al. (2017) An observation-based constraint on permafrost loss as afunction of global warming. Nat Clim Chang 7:340–344.

8. McClelland JW, et al. (2008) Development of a pan-arctic database for river chemistry.Eos (Wash DC) 89:217–218.

9. Amon RMW, et al. (2012) Dissolved organic matter sources in large Arctic rivers.Geochim Cosmochim Acta 94:217–237.

10. McClelland JW, et al. (2016) Particulate organic carbon and nitrogen export frommajor Arctic rivers. Global Biogeochem Cycles 30:629–643.

11. Guo L, Macdonald RW (2006) Source and transport of terrigenous organic matter inthe upper Yukon River: Evidence from isotope (δ13C, Δ14C, and δ15N) composition ofdissolved, colloidal, and particulate phases. Global Biogeochem Cycles 20:GB2011.

12. Striegl RG, Dornblaser MM, Aiken GR, Wickland KP, Raymond PA (2007) Carbon ex-port and cycling by the Yukon, Tanana, and Porcupine rivers, Alaska, 2001-2005.Water Resour Res 43:2001–2005.

13. Gustafsson Ö, Van Dongen BE, Vonk JE, Dudarev OV, Semiletov IP (2011) Widespreadrelease of old carbon across the Siberian Arctic echoed by its large rivers. Biogeosciences8:1737–1743.

14. Feng X, et al. (2013) Differential mobilization of terrestrial carbon pools in EurasianArctic river basins. Proc Natl Acad Sci USA 110:14168–14173.

15. Trumbore S (2009) Radiocarbon and soil carbon dynamics. Annu Rev Earth Planet Sci37:47–66.

16. Andersson A, et al. (2015) Regionally-varying combustion sources of the January 2013severe haze events over eastern China. Environ Sci Technol 49:2038–2043.

17. Sorokin YI, Sorokin PY (1996) Plankton and primary production in the Lena Riverestuary and in the south-eastern Laptev Sea. Estuar Coast Shelf Sci 43:399–418.

18. Winterfeld M, Laepple T, Mollenhauer G (2015) Characterization of particulate or-ganic matter in the Lena River delta and adjacent nearshore zone, NE Siberia–Part 1:Radiocarbon inventories. Biogeosciences 12:3769–3788.

19. Semiletov IP, et al. (2011) Carbon transport by the Lena River from its headwaters tothe Arctic Ocean, with emphasis on fluvial input of terrestrial particulate organiccarbon vs. carbon transport by coastal erosion. Biogeosciences 8:2407–2426.

20. Pipko II, Pugach SP, Dudarev OV, Charkin AN, Semiletov IP (2010) Carbonate parametersof the Lena river: Characteristics and distribution. Geochem Int 48:1131–1137.

21. Denfeld BA, Frey KE, Sobczak WV, Mann PJ, Robert M (2013) Summer CO2 evasion fromstreams and rivers in the Kolyma River basin, north-east Siberia. Polar Res 32:19704.

22. Lauerwald R, Laruelle GG, Hartmann J, Ciais P, Regnier PAG (2015) Spatial patterns inCO2 evasion from the global river network. Global Biogeochem Cycles 29:534–554.

23. Serikova S, et al. (2018) High riverine CO2 emissions at the permafrost boundary ofWestern Siberia. Nat Geosci 11:825–829.

24. Holmes RM, et al. (2012) Seasonal and annual fluxes of nutrients and organic matterfrom large rivers to the Arctic Ocean and surrounding seas. Estuaries Coasts 35:369–382.

25. Lawrence DM, Slater AG, Swenson SC (2012) Simulation of present-day and futurepermafrost and seasonally frozen ground conditions in CCSM4. J Clim 25:2207–2225.

26. Parazoo NC, Koven CD, Lawrence DM, Romanovsky V, Miller CE (2018) Detecting thepermafrost carbon feedback: Talik formation and increased cold-season respiration asprecursors to sink-to-source transitions. Cryosph 12:123–144.

27. Walvoord MA, Striegl RG (2007) Increased groundwater to stream discharge frompermafrost thawing in the Yukon River basin: Potential impacts on lateral export ofcarbon and nitrogen. Geophys Res Lett 34:L12402.

28. Limpens J, et al. (2008) Peatlands and the carbon cycle: From local processes to globalimplications. Biogeosciences 5:1475–1491.

29. Tanski G, Couture N, Lantuit H, Eulenburg A, Fritz M (2016) Eroding permafrostcoasts release low amounts of dissolved organic carbon (DOC) from ground iceinto the nearshore zone of the Arctic Ocean. Global Biogeochem Cycles 30:1054–1068.

30. Fedorova I, et al. (2015) Lena Delta hydrology and geochemistry: Long-term hydro-logical data and recent field observations. Biogeosciences 12:345–363.

31. Guo L, Ping CL, Macdonald RW (2007) Mobilization pathways of organic carbonfrom permafrost to arctic rivers in a changing climate. Geophys Res Lett 34:L13603.

32. Frey KE, McClelland JW (2009) Impacts of permafrost degradation on arctic riverbiogeochemistry. Hydrol Processes 23:169–182.

33. Galimov EM, Kodina LA, Stepanets OV, Korobeinik GS (2006) Biogeochemistry of theRussian Arctic. Kara Sea: Research results under the SIRRO project, 1995–2003. GeochemInt 44:1053–1104.

34. Tesi T, et al. (2017) Carbon geochemistry of plankton-dominated samples in theLaptev and East Siberian shelves: Contrasts in suspended particle composition. OceanSci 13:735–748.

35. Vonk JE, et al. (2015) Biodegradability of dissolved organic carbon in permafrost soilsand aquatic systems: A meta-analysis. Biogeosciences 12:6915–6930.

36. Vonk JE, et al. (2013) High biolability of ancient permafrost carbon upon thaw.Geophys Res Lett 40:2689–2693.

37. Spencer RGM, et al. (2015) Detecting the signature of permafrost thaw in Arctic rivers.Geophys Res Lett 42:2830–2835.

38. Heslop JK, et al. (2017) Variable respiration rates of incubated permafrost soil extractsfrom the Kolyma River lowlands, north-east Siberia. Polar Res 37:1305157.

39. Mann PJ, et al. (2012) Controls on the composition and lability of dissolved or-ganic matter in Siberia’s Kolyma River basin. J Geophys Res Biogeosci 117:G01028.

40. Roehm CL, Giesler R, Karlsson J (2009) Bioavailability of terrestrial organic carbon tolake bacteria: The case of a degrading subarctic permafrost mire complex. J GeophysRes Biogeosci 114:G03006.

41. Balcarczyk KL, Jones JB, Jaffe R, Maie N (2009) Stream dissolved organic matter bio-availability and composition in watersheds underlain with discontinuous permafrost.Biogeochemistry 94:255–270.

42. Kawahigashi M, Kaiser K, Kalbitz K, Rodionov A, Guggenberger G (2004) Dissolvedorganic matter in small streams along a gradient from discontinuous to continuouspermafrost. Glob Change Biol 10:1576–1586.

43. Holmes RM, et al. (2008) Lability of DOC transported by Alaskan rivers to the ArcticOcean. Geophys Res Lett 35:L09401.

44. Drake TW, Wickland KP, Spencer RGM, McKnight DM, Striegl RG (2015) Ancient low-molecular-weight organic acids in permafrost fuel rapid carbon dioxide productionupon thaw. Proc Natl Acad Sci USA 112:13946–13951.

45. Semiletov I, et al. (2016) Acidification of East Siberian Arctic Shelf waters throughaddition of freshwater and terrestrial carbon. Nat Geosci 9:361–365.

46. Gordeev VV (2006) Fluvial sediment flux to the Arctic Ocean. Geomorphology 80:94–104.

47. Lewis T, Lamoureux SF (2010) Twenty-first century discharge and sediment yieldpredictions in a small high Arctic watershed. Global Planet Change 71:27–41.

48. Steffen W, et al. (2018) Trajectories of the Earth System in the Anthropocene. ProcNatl Acad Sci USA 115:8252–8259.

49. Schuur EAG, et al. (2015) Climate change and the permafrost carbon feedback.Nature 520:171–179.

50. Vonk JE, Gustafsson Ö (2013) Permafrost-carbon complexities. Nat Geosci 6:675–676.

51. Bröder L, Tesi T, Andersson A, Semiletov I, Gustafsson Ö (2018) Bounding cross-shelftransport time and degradation in Siberian-Arctic land-ocean carbon transfer. NatCommun 9:806.

52. Raymond PA, et al. (2007) Flux and age of dissolved organic carbon exported to theArctic Ocean: A carbon isotopic study of the five largest arctic rivers. GlobalBiogeochem Cycles 21:GB4011.

53. Harrell FEJ (2018) Hmisc: Harrell Miscellaneous. R Package Version 4.1-1. Available athttps://cran.r-project.org/package=Hmisc. Accessed November 2, 2018.

54. R Core Team (2018) R: A language and environment for statistical computing.Available at https://www.r-project.org/. Accessed October 31, 2018.

55. Pasek J (2018) weights: Weighting and Weighted Statistics. R Package Version1.0. Available at https://cran.r-project.org/package=weights. Accessed Novem-ber 2, 2018.

56. HH (2018) HH: Statistical analysis and data display: Heiberger and Holland R PackageVersion 3.1-35. Available at https://cran.r-project.org/package=HH. Accessed Novem-ber 2, 2018.

57. Vonk JE, et al. (2012) Activation of old carbon by erosion of coastal and subseapermafrost in Arctic Siberia. Nature 489:137–140.

58. Tesi T, et al. (2016) Massive remobilization of permafrost carbon during post-glacialwarming. Nat Commun 7:13653.

59. Levin I, Kromer B, Hammer S (2013) Atmospheric Δ14CO2 trend in WesternEuropean background air from 2000 to 2012. Tellus B Chem Phys Meterol 65:20092.

60. Brown J, Ferrians OJ,Jr, Heginbottom JA, Melnikov ES (2002) Circum-Arctic Map of Perma-frost andGround Ice Conditions (National Snow and Ice Data Center, Boulder, CO), Version 2.

61. Strauss J, et al. (2016) Database of Ice-Rich Yedoma Permafrost (IRYP). PANGAEA,10.1594/PANGAEA.861733.

62. Wild B, Gustafsson O (2019) Content of carbon isotopes (13C and 14C) in organiccarbon pools in terrestrial Siberia. Bolin Centre Database. Available at https://bolin.su.se/data/wild-2019. Deposited April 16, 2019.

63. Wild B, Andersson A, Gustafsson O (2019) Matlab MCMC script for source appor-tionment of organic carbon in Arctic Rivers. Bolin Centre Database. Available athttps://git.bolin.su.se/bolin/wild-2019. Deposited April 23, 2019.

Wild et al. PNAS | May 21, 2019 | vol. 116 | no. 21 | 10285

ENVIRONMEN

TAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 22

, 202

0